ERIC ED564417: Validity of the SAT® for Predicting First-Year Grades: 2012 SAT Validity Sample. Statistical Report 2015 2

2015·0.59 MB·English

by  ERIC

Y

TValidity of the SAT® for Predicting First-Year Grades:
I
2012 SAT Validity Sample
D

I
L
By Jonathan Beard and Jessica P. Marini
A

V

COLLEGE BOARD RESEARCH Statistical Report 2015 -2

Executive Summary

The continued accumulation of validity evidence for the intended uses of
educational assessment scores is critical to ensure that inferences made using
the scores are sound. To that end, the College Board has continued to collect
college outcome data to evaluate the relationship between SAT® scores and
college success. This report provides updated validity evidence for using the SAT
to predict first-year college grade point average (FYGPA) for the 2012 cohort.
Colleges and universities (henceforth, “institutions”) provided data on the cohort
of first-time, first-year students enrolling in the fall of 2012. The College Board
combined those college outcomes data with official SAT scores and SAT
Questionnaire response data, which include students’ self-reported high school
grade point average among other things. In particular, 165 institutions provided
data on 300,389 students, 223,109 of whom had complete data on high school
grade point average (HSGPA); SAT critical reading (SAT-CR), mathematics
(SAT-M), and writing (SAT-W) scores; and FYGPA.
As has been shown in previous research (Kobrin, Patterson, Shaw, Mattern, &
Barbuti, 2008; Patterson, Mattern, & Kobrin, 2009; Patterson & Mattern, 2011;
2012; 2013a; 2013b), the multiple correlation of SAT section scores and HSGPA
with FYGPA continues to be strong for the 2012 cohort (r = .61). When
compared with the correlation of HSGPA alone with FYGPA (r = .53), the
addition of the SAT section scores to HSGPA represented an increase (∆r = .08)
in the correlation with FYGPA. The patterns of differential validity by institutional
and student characteristics and differential prediction by student characteristics
for the 2012 cohort also follow the same general patterns, as has been shown in
previous research (Mattern, Patterson, Shaw, Kobrin, & Barbuti, 2008; Patterson
et al., 2009; Patterson & Mattern, 2011; 2012; 2013a; 2013b.

© 2015 The College Board. 1
Table 1.
Distribution of Institutional Characteristics
Institutional Characteristic %
U.S. Region Midwest 16

Mid-Atlantic 27

New England 10

South 19

Southwest 12

West 16

Control Public 47
Private 53
Admittance Under 50% 22
Rate 50 to 75% 56
Over 75% 21
Undergraduate Small 20
Enrollment1 Medium 40
Large 20

Very Large 20
Note: Number of institutions (K) = 165. Percentages may not sum to 100 due to rounding.
Undergraduate enrollment was categorized as follows: small: 750 to 1,999; medium:
2,000 to 7,499; large: 7,500 to 14,999; and very large: 15,000 or more.

• Table 1 shows that the sample of 165 four-year institutions was diverse with respect to region of the U.S., control, size, and undergraduate
admittance rate.

Table 2.
Descriptive Statistics for Total Sample
Variable M SD
HSGPA 3.62 0.496
SAT-CR 550 97.4
SAT-M 571 99.7
SAT-W 544 99.5
FYGPA 3.02 0.718
Note: Number of students (N) = 223,109.

• When compared to the 2011 cohort, Table 2 shows similar mean performance for high school grade point average (HSGPA) and first-year
grade point average (FYGPA) for the 2012 cohort, with means differing by at most 0.02. The mean SAT section scores are also quite similar,
differing by no more than one point for any one section (Patterson & Mattern, 2013b).
• When compared with the population of all college-bound SAT takers expecting to graduate in 2012 (n = 1,664,479) — whose mean and
(standard deviation) SAT critical reading (SAT-CR), mathematics (SAT-M), and writing scores (SAT-W) were 496 (114), 514 (117), and
488 (113), respectively (College Board, 2012) — the sample in this study performed better in terms of SAT section scores. Given that
students in this subsample not only chose to take the SAT — as did the population of over 1.6 million college-bound seniors — but also
applied to, enrolled at, and earned grades at a four-year institution, their higher mean SAT section scores followed the expected pattern.

1. Seven institutions were missing values for Undergraduate Enrollment.

research.collegeboard.org © 2015 The College Board. 2
Table 3.
Corrected (Raw) Correlation Matrix of SAT, HSGPA, and FYGPA
Variable HSGPA SAT-CR SAT-M SAT-W FYGPA
HSGPA .45 .48 .48 .53
SAT-CR (.20) .72 .84 .48
SAT-M (.22) (.49) .74 .48
SAT-W (.23) (.69) (.51) .52
FYGPA (.34) (.27) (.26) (.33)
Note: Number of students (N) = 223,109. Pooled within institution, restriction of range corrected correlations are presented. The raw
correlations are shown in parentheses.

• Table 3 shows the restriction of range corrected and raw correlations among the four predictors examined in this study: HSGPA, SAT-CR,
SAT-M, and SAT-W, as well as FYGPA. In general, SAT sections were more highly correlated with other sections than with HSGPA, and this is
most prominent in the correlation of SAT-CR and SAT-W.
• The bivariate correlations shown in Table 3 are similar to what was estimated in previous research (Kobrin et al., 2008; Patterson et al., 2009;
Patterson & Mattern, 2011; Patterson & Mattern, 2012; Patterson & Mattern, 2013a, 2013b).
• Consistent with prior research, the SAT writing section had the highest correlation with FYGPA among the three SAT section scores (Kobrin
et al., 2008; Patterson et al., 2009; Patterson & Mattern, 2011; Patterson & Mattern, 2012; Patterson & Mattern, 2013a, 2013b).

Table 4.
Corrected (Raw) Multiple Correlations of Predictors with FYGPA
Predictor(s) Correlation
1. SAT-M, SAT-CR .51 (.31)
2. HSGPA, SAT-M, SAT-CR .60 (.42)
3. SAT-CR, SAT-M, SAT-W .54 (.35)
4. HSGPA, SAT-CR, SAT-M, SAT-W .61 (.44)
Note: Number of students (N) = 223,109. Multiple correlations are based on the raw and corrected correlations presented in Table 3.
The raw correlations are shown in parentheses.

• SAT-CR, SAT-M, and SAT-W jointly have a similar multiple correlation with FYGPA (.54) as does HSGPA with FYGPA (.53). It is, however, the
inclusion of all four predictors that leads to the strongest linear relationship with FYGPA; namely, a multiple correlation of .61.

research.collegeboard.org © 2015 The College Board. 3
Figure 1.
Mean FYGPA by SAT score band.
0
0
4.
0 3.54
5
3. 3.30
0 3.02
0
3.
A 2.67
GP 0
FY 2.5 2.38
0
0
2.
0
5
1.
0
0
1. 600-1190 1200-1490 1500-1790 1800-2090 2100-2400
SAT Score Band

Note: SAT score bands are based on the sum of SAT-CR, SAT-M, and SAT-W.
Sample sizes by SAT score band were as follows:

SAT n
600–1190 8,239

1200–1490 52,628

1500–1790 90,234

1800–2090 60,054

2100–2400 11,954

• Figure 1 shows graphically the positive relationship between the composite SAT score band (i.e., sum of SAT-CR, SAT-M, and SAT-W,
grouped into meaningful categories) with mean FYGPA. In particular, the difference in mean FYGPA between the highest score band
(2100–2400) and the lowest (600–1190) was 1.16. In other words, students in the highest SAT score band earned, on average, an FYGPA of
A- , compared to students in the lowest SAT score band, who had an average FYGPA of C+.

research.collegeboard.org © 2015 The College Board. 4
Figure 2.
Percentage of students earning FYGPA of a B or higher by SAT score band.

0
0
1
88%
0 77%
8
age 60 59%
t
n
e
c
r
e
P 0 38%
4
23%
0
2
0
600-1190 1200-1490 1500-1790 1800-2090 2100-2400
SAT Score Band

Note: SAT score bands are based on the sum of SAT-CR, SAT-M, and SAT-W.
Students whose FYGPA was ≥ 3.00 were considered to have earned a B or better.
Sample sizes by SAT score band were as follows:

SAT n
600–1190 8,239

1200–1490 52,628

1500–1790 90,234

1800–2090 60,054

2100–2400 11,954

• Figure 2 shows graphically the positive relationship between the percentage of students earning at least a B (i.e., 3.0 FYGPA or higher) in
their first year of college with the composite SAT score band. In particular, over three and a half times the number of students in the highest
SAT score band (2100–2400) earned at least a B, relative to those in the lowest (600–1190).

research.collegeboard.org © 2015 The College Board. 5
Figure 3.
Incremental validity of the SAT: Mean FYGPA by SAT score band controlling for HSGPA.

4
3.57
5
3. 3.37
3.17 3.14 3.15
2.98
3 2.87
2.79
A 2.72
P 2.56 2.58
G 5 2.43
Y 2. 2.36
F 2.22
2.08
2
5
1.
1
C or Lower B A
HSGPA
SAT Score Ranges
0600-1190 1200-1490 1500-1790 1800-2090 2100-2400

Note: SAT score bands are based on the sum of SAT-CR, SAT-M, and SAT-W.
HSGPA ranges were defined as follows:

“A” range: 4.33 (A+), 4.00 (A), and 3.67 (A-);

“B” range: 3.33 (B+), 3.00 (B), and 2.67 (B-); and

“C or Lower” range: 2.33 (C+) or lower.

Sample sizes by HSGPA and SAT score band were as follows:

HSGPA

SAT C or Lower B A

600–1190 1,040 5,208 1,991

1200–1490 2,685 28,234 21,709

1500–1790 1,371 29,061 59,802

1800–2090 340 9,754 49,960

2100–2400 32 935 10,987

• Figure 3 shows the relationship of composite SAT score band with mean FYGPA at different levels of HSGPA. For each level of HSGPA,
higher composite SAT score bands are associated with higher mean FYGPAs, thus demonstrating the value of SAT above and beyond
HSGPA in the prediction of FYGPA.
• Consider, for example, two students with HSGPAs in the A range; the one whose SAT composite was 600–1190 was expected to earn an
FYGPA of 2.6, which translates to roughly a B-, while the other student, whose SAT composite was 2100–2400, was expected to earn an
FYGPA of 3.6, which translates to roughly an A-.

research.collegeboard.org © 2015 The College Board. 6
Table 5.
Descriptive Statistics of Study Variables by Institutional Characteristics
SAT-CR SAT-M SAT-W HSGPA FYGPA
Institutional Characteristic k n M SD M SD M SD M SD M SD
Control Private 87 62,391 575 97.8 592 98.8 575 100.9 3.66 0.479 3.17 0.591
Public 78 160,718 540 95.4 563 98.8 532 96.3 3.61 0.502 2.96 0.754
Admittance Under 50% 37 51,047 600 91.9 625 90.2 603 95.1 3.79 0.406 3.21 0.551
Rate 50 to 75% 93 144,808 540 93.3 562 96.0 532 93.7 3.60 0.497 2.99 0.738
Over 75% 35 27,254 507 92.7 516 91.2 497 89.0 3.40 0.535 2.80 0.806
Undergraduate Small 32 9,527 540 102.8 544 98.5 531 100.3 3.54 0.532 3.02 0.679
Enrollment2 Medium 63 42,757 545 102.5 557 103.2 540 105.4 3.53 0.535 3.03 0.738
Large 31 56,122 547 97.7 566 102.1 541 101.1 3.59 0.505 3.01 0.713

Very Large 32 104,506 555 93.7 581 95.2 548 95.6 3.68 0.467 3.02 0.724
Total 165 223,109 550 97.4 571 99.7 544 99.5 3.62 0.496 3.02 0.718
Note: k: number of institutions, n: subgroup sample size. Undergraduate enrollment was categorized as follows: small: 750 to 1,999;
medium: 2,000 to 7,499; large: 7,500 to 14,999; and very large: 15,000 or more.

• Table 5 provides summary statistics on the key study variables by institutional characteristics.
• It shows that, in general, mean SAT section scores, HSGPA, and FYGPA were higher:
at private institutions, compared to public institutions; and
o
at increasingly selective institutions (i.e., those that admit fewer applicants).
o
• In terms of undergraduate enrollment (i.e., institution size):
small institutions had the lowest mean SAT section scores, while very large institutions had the highest mean SAT section scores and
o
HSGPAs; and
there is no discernable relationship between institution size and mean FYGPA.
o

2. Seven institutions (10,197 students) were missing values for Undergraduate Enrollment.

research.collegeboard.org © 2015 The College Board. 7
Table 6.
Corrected Correlations of SAT and HSGPA with FYGPA by Institutional Characteristics
Institutional Characteristic k n SAT-CR SAT-M SAT-W SAT* HSGPA SAT*, HSGPA
Control Private 87 62,391 .53 .52 .57 .59 .56 .66
Public 78 160,718 .45 .46 .50 .52 .52 .60
Admittance Under 50% 37 51,047 .52 .52 .56 .58 .53 .64
Rate 50 to 75% 93 144,808 .46 .47 .51 .53 .53 .61
Over 75% 35 27,254 .46 .46 .51 .52 .54 .61
Undergraduate Small 32 9,527 .56 .55 .60 .62 .59 .70
Enrollment3 Medium 63 42,757 .48 .48 .53 .54 .55 .63
Large 31 56,122 .48 .49 .52 .55 .54 .62
Very Large 32 104,506 .46 .47 .51 .52 .52 .60
Overall 165 223,109 .48 .48 .52 .54 .53 .61
Note: k: number of institutions, n: subgroup sample size. SAT* refers to the inclusion of all three sections in the relevant multiple
correlation. The correlations were corrected for restriction of range within institutions and pooled. Undergraduate enrollment was
categorized as follows: small: 750 to 1,999; medium: 2,000 to 7,499; large: 7,500 to 14,999; and very large: 15,000 or more. For
raw correlations by institutional characteristics, see Appendix B.

• Table 6 shows the correlations of various combinations of the predictors with FYGPA by key institutional characteristics.
• The correlations of the six combinations of predictors with FYGPA are generally:
Higher at private than public institutions
o
Highest at the most selective institutions (i.e., those that admit fewer than 50% of applicants), relative to less selective ones
o
Highest at small institutions, relative to larger ones
o
• Across many institutional categories, the multiple correlation of SAT with FYGPA was at least as high as the correlation of HSGPA with
FYGPA.
• For nearly all institutional subgroups, the combination of SAT section scores and HSGPA represented an increase in at least .07 for the
correlation with FYGPA over either predictor alone.

3. Seven institutions were missing values for Undergraduate Enrollment.

research.collegeboard.org © 2015 The College Board. 8
Table 7.
Descriptive Statistics of Study Variables by Student Characteristics
SAT-CR SAT-M SAT-W HSGPA FYGPA
Student Characteristic k n M SD M SD M SD M SD M SD
Gender Male 162 100,739 557 97.0 595 98.3 541 100.1 3.58 0.511 2.93 0.754
Female 165 122,370 544 97.3 551 96.5 547 98.9 3.66 0.481 3.10 0.678
Racial / African American 165 19,326 483 89.2 486 87.6 470 88.8 3.37 0.555 2.62 0.798
Ethnic American Indian 145 991 540 91.2 554 92.6 526 92.1 3.57 0.513 2.85 0.774
Identity Asian 164 25,399 567 105.4 637 99.3 580 108.1 3.70 0.448 3.13 0.641
Hispanic 165 24,787 507 96.4 525 95.2 502 94.2 3.55 0.512 2.82 0.752
Other 163 6,135 560 99.2 572 100.9 556 102.5 3.59 0.503 3.00 0.718
White 165 144,464 563 90.9 578 90.8 554 92.9 3.66 0.480 3.09 0.688
Not Stated 163 2,007 546 102.4 555 103.6 535 104.2 3.53 0.545 2.93 0.735
Best English Only 165 190,113 555 95.2 571 97.0 546 97.7 3.63 0.495 3.03 0.717
Language English and Another 165 28,411 525 102.6 559 109.6 530 106.6 3.59 0.501 2.92 0.727
Another Language 150 3,856 478 103.3 642 118.2 521 116.8 3.67 0.480 3.10 0.659
Not Stated 140 729 543 104.6 558 109.2 529 108.3 3.51 0.631 2.96 0.725
Household < $40,000 165 19,820 503 96.5 527 103.0 494 96.4 3.54 0.540 2.78 0.809
Income $40,000–80,000 165 25,308 536 93.9 553 96.5 524 94.5 3.62 0.509 2.93 0.765
$80,000–120,000 165 24,714 554 92.1 574 94.2 544 94.6 3.65 0.497 3.04 0.710
$120,000–160,000 165 12,199 562 91.3 583 92.9 555 93.9 3.65 0.488 3.07 0.678
$160,000–200,000 165 6,696 569 92.2 592 93.3 566 94.0 3.63 0.495 3.11 0.652
> $200,000 162 12,516 586 88.5 612 89.6 588 93.3 3.64 0.475 3.15 0.609
Not Stated 165 121,856 554 97.9 574 99.9 549 99.6 3.63 0.488 3.05 0.702
Highest No High School Diploma 161 7,314 470 90.9 512 103.7 470 90.0 3.50 0.526 2.76 0.780
Parental High School Diploma 165 44,289 512 90.0 532 95.3 503 91.3 3.55 0.519 2.83 0.786
Education Associate Degree 165 14,802 520 87.6 537 90.9 507 88.7 3.58 0.516 2.89 0.765
Level Bachelor’s Degree 165 78,556 558 89.4 580 92.3 552 91.7 3.66 0.479 3.08 0.677
Graduate Degree 165 65,745 591 93.3 608 94.5 587 95.9 3.70 0.464 3.17 0.637
Not Stated 165 12,403 500 95.2 525 100.5 498 96.5 3.41 0.526 2.81 0.765
Total 165 223,109 550 97.4 571 99.7 544 99.5 3.62 0.496 3.02 0.718
Note: n: subgroup sample size.

• Table 7 shows that female students tended to outperform males on SAT-W, HSGPA, and FYGPA, while the opposite was true for SAT-CR and
SAT-M.
• Some differences exist across racial/ethnic identities, with white and Asian students having higher mean SAT section scores, HSGPA, and
FYGPA relative to African American and Hispanic students.
• When considering best spoken language, students whose best language was English had the highest SAT-CR and SAT-W scores, whereas
students whose best language was something other than English had the highest SAT-M scores and slightly higher HSGPA and FYGPA
means.
• Students who reported higher household incomes had higher mean SAT section scores and FYGPA, but there was no apparent relationship
with HSGPA.
• As with household-income level, mean SAT section scores and FYGPA increased as highest parental education level increased; with respect
to mean HSGPA, there was a positive relationship with highest parental education level.

research.collegeboard.org © 2015 The College Board. 9
Table 8.
Corrected Correlation of SAT Scores and HSGPA with FYGPA by Student Subgroups
Student Characteristic k n SAT-CR SAT-M SAT-W SAT* HSGPA SAT*, HSGPA
Gender Male 162 100,739 .46 .48 .50 .52 .52 .60
Female 165 122,370 .52 .53 .55 .58 .53 .64
Racial / African American 134 19,033 .42 .40 .46 .47 .44 .52
Ethnic American Indian 17 437 .33 .32 .39 .40 .36 .44
Identity Asian 121 25,066 .45 .48 .48 .52 .49 .58
Hispanic 139 24,566 .43 .42 .47 .48 .46 .54
Other 102 5,671 .44 .46 .48 .51 .49 .57
White 165 144,464 .47 .47 .52 .54 .55 .63
Not Stated 51 1,336 .36 .38 .41 .42 .44 .50
Best English Only 165 190,113 .48 .48 .53 .55 .54 .63
Language English and Another 152 28,291 .43 .43 .47 .48 .46 .54
Another Language 52 3,336 .38 .41 .41 .44 .40 .48
Not Stated 8 156 .34 .40 .43 .45 .29 .46
Household < $40,000 156 19,732 .40 .41 .44 .46 .45 .52
Income $40,000–80,000 162 25,268 .45 .46 .49 .51 .52 .59
$80,000–120,000 162 24,682 .46 .46 .50 .52 .54 .61
$120,000–160,000 136 11,927 .45 .45 .50 .51 .55 .61
$160,000–200,000 105 6,293 .45 .45 .49 .51 .55 .61
> $200,000 107 12,119 .47 .45 .50 .52 .55 .61
Not Stated 165 121,856 .49 .49 .53 .55 .54 .63
Highest No High School Diploma 74 6,712 .41 .45 .44 .48 .43 .52
Parental High School Diploma 162 44,257 .43 .44 .48 .50 .49 .57
Education Associate Degree 144 14,634 .44 .45 .50 .51 .51 .59
Level Bachelor’s Degree 165 78,556 .47 .47 .51 .53 .54 .62
Graduate Degree 163 65,718 .49 .49 .53 .55 .56 .64
Not Stated 142 12,220 .41 .41 .45 .46 .46 .53
Overall 165 223,109 .48 .48 .52 .54 .53 .61
Note: k: number of institutions, n: subgroup sample size. SAT* refers to the inclusion of all three sections in the relevant multiple
correlation. The correlations were corrected for restriction of range within institutions and pooled. Institutions with fewer than 15
students in any subgroup were excluded. For raw correlations by institutional characteristics, see Appendix C.

• Table 8 shows that predictive validity for all predictors and combinations of FYGPA was higher for:
Female students than for male students;
o
White and Asian students relative to the other racial/ethnic identities;
o
Students whose best spoken language was English only as compared to the other language groups;
o
Students with a household income level of at least $80,000 than those with lower incomes; and
o
Students whose parents have higher education levels as compared to lower education levels.
o
• Across all student subgroups, predictive validity of FYGPA was maximized using the combination of SAT section scores and HSGPA.

research.collegeboard.org © 2015 The College Board. 10